Antisaccade performance in biological relatives of schizophrenia patients: a meta-analysis

Schizophrenia Research 71 (2004) 113 – 125 www.elsevier.com/locate/schres

Antisaccade performance in biological relatives of schizophrenia patients: a meta-analysis Deborah L. Levy a,*, Gillian O’Driscoll b, Steven Matthysse a, Samantha R. Cook c, Philip S. Holzman a, Nancy R. Mendell d a Psychology Research Laboratory, McLean Hospital, 115 Mill Street, Belmont, MA 02478, USA Department of Psychology and Douglas Hospital Research Center, McGill University, Montreal, Quebec, Canada c Department of Statistics, Harvard University, Cambridge, MA 02138, USA d Department of Applied Mathematics and Statistics, State University of New York at Stony Brook, Stony Brook, NY 11794, USA b

Received 4 September 2003; accepted 12 November 2003 Available online 22 January 2004

Abstract Poor performance on the antisaccade (AS) task has been interpreted as a potential indicator of genetic liability that may enhance the power of linkage studies of a multidimensional phenotype for schizophrenia. Every study has replicated the finding of significantly worse performance in schizophrenia patients regardless of which specific antisaccade paradigm was employed. In some studies involving a standard version of the antisaccade task, relatives of schizophrenia patients made an increased number of errors, but in other studies that used this same paradigm, relatives of schizophrenia patients did not differ from controls. In this paper, we report the results of a meta-analysis on studies that used the standard antisaccade paradigm. The meta-analysis shows that those studies that reported large effect sizes and statistically significant differences between relatives of schizophrenia patients and controls used inclusion/exclusion criteria that were not symmetrical between the two groups, whereas those studies that reported small and nonsignificant differences between relatives of schizophrenia patients and controls used symmetrical inclusion/exclusion criteria. Specifically, studies that applied stricter psychopathology exclusion criteria to controls than to relatives of schizophrenia patients had larger effect sizes than studies that applied comparable exclusion criteria to both groups, suggesting that antisaccade performance is compromised by psychopathology in general rather than by schizophrenia per se. Since symmetrical inclusion/exclusion criteria between relatives of schizophrenia patients and controls are essential for a genetic analysis, and those studies that did apply symmetrical criteria had small effect sizes, the available data suggest that poor antisaccade performance is unlikely to be useful in identifying clinically unaffected carriers of genes for schizophrenia. D 2003 Elsevier B.V. All rights reserved. Keywords: Schizophrenia; Antisaccade performance; Meta-analysis

1. Introduction * Corresponding author. Tel.: +1-617-855-2854; fax: +1-617855-2778. E-mail address: [email protected] (D.L. Levy). 0920-9964/$ - see front matter D 2003 Elsevier B.V. All rights reserved. doi:10.1016/j.schres.2003.11.006

The risk for schizophrenia in first-degree biological relatives of schizophrenics (RSP) is only about 6.5% (Kendler et al., 1993), possibly too low to have

114

D.L. Levy et al. / Schizophrenia Research 71 (2004) 113–125

adequate power to detect linkage even when it is present. Traits that are substantially more penetrant than schizophrenia in RSP could increase the power of linkage studies (Matthysse and Parnas, 1992; see Greenberg et al., 2000, for the usefulness of such traits in relation to genetic studies of nonpsychiatric disorders). The finding of increased errors on the antisaccade (AS) task not only in schizophrenia patients, but also in some studies of RSP, has led to the speculation that poor AS performance may tap processes related to genetic vulnerability (Clementz et al., 1994; Katsanis et al., 1997; McDowell and Clementz, 1997; Crawford et al., 1998; McDowell et al., 1999; Thaker et al., 2000; Karoumi et al., 2001; Curtis et al., 2001a; Ross et al., 1998). The AS task requires the subject, who is fixating a central target, to inhibit a saccade to an abrupt-onset peripheral stimulus and to generate a voluntary saccade to the mirror location in the opposite periphery, where there is no visible target. Correct saccades away from the target are called ‘antisaccades’. Saccades toward the peripheral target are considered errors. Three AS paradigms have been used in studies of RSP. In the most widely used version of the AS task (which we refer to as the ‘‘standard’’ paradigm), offset of the central fixation point and onset of the peripheral target occur simultaneously. Other versions, which have been used in fewer studies, are the ‘‘overlap’’ and the ‘‘gap’’ paradigms. In the ‘‘overlap’’ paradigm, the central fixation point stays on for a short time after the peripheral target has been illuminated. In the ‘‘gap’’ paradigm, the offset of the central fixation point precedes the appearance of the peripheral target by a short time. In any of these paradigms, peripheral targets may appear at single (e.g., F 5j) or multiple (e.g., F 5j, 10j, 15j) eccentricities and timing parameters can be fixed or variable. In this paper, we review the results of studies on RSP from each of the various AS paradigms. Since paradigm variations affect error rate and other measures (e.g., latency) (Fischer and Weber, 1992, 1997), the results of studies of RSP on the AS task are discussed for each of the three paradigms separately. We report here the results of a metaanalysis of studies of RSP that used the standard AS paradigm, including a test for heterogeneity in effect size and planned contrasts to examine the effects of

possible moderator variables on variability in effect size. We also test for heterogeneity in effect size in studies of RSP that used the overlap AS paradigm. In addition, we examine the extent to which the data are consistent with a genetic model of cofamilial transmission.

2. Results 2.1. Standard AS task 2.1.1. Magnitude of effect Nine studies compared the performance of RSP and nonpsychiatric controls on a standard version of the AS task. In five of these studies, RSP showed a significantly higher error rate than controls did (Clementz et al., 1994; Katsanis et al., 1997; McDowell and Clementz, 1997; Curtis et al., 2001a; Karoumi et al., 2001), and in four studies RSP did not differ from controls in mean error rate (Thaker et al., 1996, 2000; Crawford et al., 1998; Brownstein et al., 2003). Table 1 presents descriptive statistics as well as effect sizes for the differences between RSP and controls, calculated in three ways. Regardless of the method of calculation, the effect sizes vary considerably across studies. Using Cohen’s method [d=(meanRSP meanCONTROLS)/pooled standard deviation], d ranges from 0.05 to 0.84, consistent with effects ranging from small ( V 0.2) to large (0.8) (Cohen, 1977). The mean d is 0.43 (S.D.: 0.32) and the median is 0.51 [95% confidence intervals (CIs): 0.19 – 0.68]. The correlations between group membership and performance, r, range from a low of 0.025 to a high of 0.39, a positive correlation indicating a higher error rate in RSP than in controls. The mean value of r is 0.20 (S.D.: 0.15) and the median is 0.25 (95% CIs: 0.09– 0.32), again consistent with small to medium effects (Rosenthal and Rosnow, 1984). The effect size d assumes equal variances in the groups being compared (Cohen, 1977), but in some of the studies of AS performance in RSP, the variances were not equal. The value of d is under- or overestimated in such cases, because the denominator in the calculation of d is the pooled standard deviation of the two groups. In order to take into account unequal variances, we also calculated Glass’s delta, in which

D.L. Levy et al. / Schizophrenia Research 71 (2004) 113–125

115

Table 1 Antisaccade studies (standard paradigm) of first-degree relatives of schizophrenia patients ordered by effect size (Cohen’s d) Study

First-degree relatives of schizophrenia patients N

Brownstein et al. (2003) Thaker et al. (1996)b Thaker et al. (2000)b Crawford et al. (1998) Clementz et al. (1994) McDowell and Clementz (1997) Curtis et al. (2001a) Karoumi et al. (2001) Katsanis et al. (1997)

% Error Scorea

Nonpsychiatric controls N

% Error Scorea

Effect size Cohen’s d

r

Glass’ delta

98

23.2 (18.5)

24

24.2 (19.1)

0.05

0.025

0.05

26

26.7 (10.8)

68

24.3 (19.2)

0.14

0.07

0.13

55

28.96 (23.2)

62

24.98 (18.7)

0.19

0.095

0.21

50

33.0 (29.0)

38

27.0 (23.0)

0.23

0.114

0.26

32

32.1 (23.2)

33

22.0 (15.9)

0.51

0.25

0.63

60

17.0 (21.0)

32

7.0 (16.0)

0.51

0.25

0.625

116

38.2 (22.3)

109

24.6 (17.0)

0.68

0.32

0.80

21

36.6 (24.1)

21

20.4 (13.6)

0.83

0.38

1.20

55

45.0c (26.2d)

38

25.0c (19.5d)

0.84

0.39

1.00

Mean (standard deviation). a Error rate was age-adjusted in some studies but not in others. b Regardless of the method of calculation, all of the effect sizes in the Thaker et al. (1996, 2000) studies become smaller when the controls are restricted to community subjects without schizophrenia spectrum personality (SSP) traits. Thus, including all community subjects in the control group in the effect sizes shown above did not mask a larger difference between RSP and controls. c Median; for estimating effect size, the median was considered equivalent to the mean. d Estimated based on interquartile range as follows: (3/4)( Q3 Q1), where Q3 and Q1 were the upper and lower ends of the interquartile range, respectively.

the denominator is the standard deviation of the control group (Glass, 1976; see Rosenthal, 1984). As Table 1 shows, Glass’s delta tends to be larger than d, especially in those studies in which RSP and controls differed in variance. The mean Glass’ delta is 0.53 (S.D.: 0.42) and the median is 0.625 (95% CIs: 0.21 –0.86). One-sample t-tests on the values of Cohen’s d, r, and Glass’ delta for the various studies indicated that that each of the means differed significantly from 0 (Ps < 0.005). Although this result would seem to rule out the possibility of no difference in AS performance between RSP and controls, such an interpretation is clouded not only because there is significant heterogeneity among the studies, but also because the heterogeneity is traceable to a specific methodological feature of the studies (see next two sections below).

Therefore, the mean effect sizes are not interpretable as a population indicator, because the values differ significantly among the studies depending on the methods used and the overall mean of the group of studies will depend on the number of studies using each method. 2.1.2. Heterogeneity analysis We evaluated whether the effect sizes could be considered consistent with each other, or were contradictory, by performing an analysis for heterogeneity in effect size (Rosenthal and Rosnow, 1984). The results showed evidence of significant heterogeneity (X2 = 19.68, = 8, 0.01 < P < 0.02), indicating that the effect sizes among the various studies are not consistent with a single mean and thus were drawn from more than one distribution. Having found significant

116

D.L. Levy et al. / Schizophrenia Research 71 (2004) 113–125

heterogeneity, we tried to identify possible sources of this heterogeneity. 2.1.3. Sources of heterogeneity Based on the published literature, we identified several possible sources of heterogeneity: (1) whether symmetrical or asymmetrical diagnostic exclusion criteria were applied to controls and RSP; (2) whether relatively few or many AS trials were administered ( V 24 vs. >25, respectively); and (3) whether the peripheral target appeared at one or multiple eccentricities (1 vs. >1 eccentricity). Studies were classified as applying symmetrical diagnostic exclusion criteria if the same exclusion criteria were applied to both groups, and as applying asymmetrical exclusion criteria if the exclusion criteria differed for the two groups. The specific details of the classifications are described in footnote b of Table 2, but an example of the kinds of asymmetry that we observed in the studies being reviewed here would be excluding individuals with a history of a nonpsychotic mood disorder from the control group but not excluding them from the RSP group. To determine whether any of these variables was related to variability in the magnitude of the observed effect sizes, we performed three planned linear contrasts. In all of the contrasts, the sum of the weights for each contrast was 0 and the effect size used was d (Rosenthal and Rubin, 1982). For each contrast the nine studies could be divided into two groups of approximately equal size (i.e., 4 vs. 5). Table 2 presents the weights assigned to each study for each planned contrast. Each contrast yielded a standard score, Z, which corresponds to a p-value indicating whether a particular methodological variable accounts for significant variability in the magnitude of the observed effect sizes. Studies that applied more restrictive diagnostic exclusion criteria to controls than to RSP had a significantly larger mean effect size than studies that applied comparable exclusion criteria to both groups (Z = 3.60; P = 0.00034). Specific features of the AS procedure did not account for variability in effect size across studies: neither the contrast for number of trials (Z = 0.16; P = 0.88) nor the contrast for number of eccentricities (Z = 0.76, P = 0.44) was statistically significant.

2.2. Overlap AS task Two groups of investigators compared RSP and controls on the overlap AS task (McDowell and Clementz, 1997; McDowell et al., 1999; Curtis et al., 2001b). Because data were also available from subsets of some of these samples on the standard AS paradigm, the effect sizes for the two paradigms can be compared as well. Table 3 presents descriptive statistics and effect sizes for those studies that used the overlap paradigm. In the McDowell and Clementz (1997) study, RSP made significantly more errors than controls in both AS paradigms. The effect size, Glass’ delta, was larger for the overlap paradigm than for the standard paradigm (0.73 vs. 0.625, respectively), although Cohen’s d (0.39 vs. 0.51, respectively) and r (0.19 vs. 0.25, respectively) were actually smaller. In the Curtis et al. (2001b) study, RSP made significantly more errors than controls on the standard AS task, but not on the overlap AS task. All effect sizes for the overlap paradigm were smaller than those for the standard paradigm (Glass’ delta: 0.29 vs. 0.50, respectively; d: 0.23 vs. 0.52, respectively; r: 0.11 vs. 0.25, respectively). In the McDowell et al. (1999) study, only the overlap paradigm was used, so the relative sensitivity of the overlap vs. standard paradigms cannot be addressed. That study evaluated the effect of target eccentricity and found that the effect sizes for ‘‘far’’ overlap targets were much larger than those for ‘‘near’’ overlap targets in each of three samples of RSP.1 1 Although the overlap condition increased effect size relative to the standard condition in one study (McDowell and Clementz, 1997) and reduced it in another (Curtis et al., 2001b), the results of the two studies may, nevertheless, be consistent with each other. The single eccentricity ( F 10j) target used by Curtis et al. (2001b) is comparable to the ‘‘near’’ target ( F 8j) used by McDowell et al. (1999). The results of both the McDowell et al. (1999) and Curtis et al. (2001b) studies indicate that ‘‘near’’ overlap targets do not seem to optimize differences between RSP and controls. Thus, the overlap condition may not have produced a larger effect than the standard condition in the Curtis et al. (2001b) study because a ‘‘far’’ peripheral target may be required to detect it, and that study used only a ‘‘near’’ peripheral target. Averaging across near and far targets in the McDowell and Clementz (1997) study may have obscured the difference in effect size between the near and far overlap targets that became apparent when the eccentricities were evaluated separately in their 1999 study. This interpretation is consistent with the finding that an overlap paradigm that used only a ‘‘near’’ target normalized errors in schizophrenia patients (Levy et al., 1998).

D.L. Levy et al. / Schizophrenia Research 71 (2004) 113–125

117

Table 2 Relevant statistical values and contrast weights for planned contrast calculations: standard antisaccade paradigm ordered as in Table 1a Study

df

t

Contrast weights Symmetric vs. asymmetric diagnostic exclusion criteriab

Brownstein et al. (2003) Thaker et al. (1996) Thaker et al. (2000) Crawford et al. (1998) Clementz et al. (1994) McDowell and Clementz (1997) Curtis et al. (2001a) Karoumi et al. (2001) Katsanis et al. (1997)

120 92 115 86 63 90 223 40 91

0.27 0.67 1.02 1.07 2.02 2.42 5.08 2.62 4.01

1.0 1.0 1.0 1.0 0.80 0.80 0.80 0.80 0.80

Number of eccentricitiesc 1.25 1.0 1.0 1.0 1.25 1.0 1.25 1.0 1.25

Number of trialsd 1.0 1.25 1.25 1.0 1.0 1.25 1.0 1.25 1.0

a

Based on the information in published articles, supplemented by additional information from the authors when available. Symmetric diagnostic exclusion criteria applied to RSP and controls ( 1.0) vs. asymmetric inclusion criteria applied to RSP and controls (0.80). Specifics are as follows: symmetric criteria: Thaker et al. (1996): both groups: no personal Axis I disorder other than one episode of past major depression (none within the preceding 2 years), for which no hospitalization, tricyclic antidepressant or electroshock treatment was received; no substance abuse within 2 years; both controls and RSP included two subgroups: subjects who met subthreshold criteria (less one criterion) for schizotypal, schizoid, or paranoid personality disorder, and subjects who did not meet subthreshold criteria for these disorders; normal controls (NC): no family history of major psychosis. Crawford et al. (1998): both groups: no substance abuse or heavy alcohol use within 1 year; RSP: recurrent major depressive disorder (N = 5), bulimia (N = 1), schizotypal personality disorder (SPD) (N = 3); NC: no personal or family history of psychotic illness. Thaker et al. (2000): both groups: no personal Axis I disorder; both controls and RSP included subjects who met subthreshold criteria for schizophrenia-related personality disorders as in Thaker et al. (1996); NC: no family history of psychosis. Brownstein et al. (2003): both groups: no psychotic disorder, no schizotypal, schizoid or paranoid personality disorder, no current or past substance abuse or dependence within 1 year; RSP: nonpsychotic affective disorders (N = 26), anxiety disorders (N = 6), substance use disorders (N = 18), adjustment disorder (N = 1); NC: no family history of psychosis; nonpsychotic affective disorders (N = 9), anxiety disorders (N = 3), substance use disorders (N = 4). Asymmetric criteria: Clementz et al. (1994): RSP: schizophrenia (N = 1), SPD (N = 3), past major depressive disorder (N = 7); NC: no major affective disorder, psychotic disorder, or current psychoactive substance use disorder; no schizophrenia-related personality disorder; no T score >70 on MMPI-2 scales L, F, 2, 6, 7, 8; no family history of psychotic disorder, suicide or psychiatric hospitalization. McDowell and Clementz (1997): RSP: schizophrenia (N = 1); morbid risk rate for schizophrenia: 2.3%; ‘‘Relatives were given SCID diagnoses’’, but no other diagnostic information about specific disorders in RSP was included; NC: no Axis I disorder, no T score >70 on MMPI-2 scales L, F, 2, 6, 7, 8; no family history of psychotic disorder, suicide or psychiatric hospitalization. Katsanis et al. (1997): RSP: schizophrenia (N = 4), past or current major depressive disorder without psychotic features (N = 5), past bipolar disorder without psychotic features (N = 1), past bipolar disorder with psychotic features (N = 1), psychotic disorder not otherwise specified (N = 1); past substance abuse or dependence (N = 5), current substance dependence and past major depressive disorder without psychotic features (N = 1); NC: no personal or family history of major affective, psychotic or substance use disorder. Curtis et al. (2001a): RSP: past but not current psychotic disorder (schizophrenia, bipolar disorder, delusional disorder) (N = 8); nonpsychotic Axis I disorders (depression, substance dependence) (N = 36); NC: no mood disorder, psychotic symptoms, lifetime substance dependence, or current substance abuse; no personal or first-degree family history of treatment for any psychiatric disorder. Karoumi et al. (2001): both groups: no Axis I disorder; no SPD; NC: no first-degree family history of Axis I disorder. c Eccentricity: one ( 1.25) vs. >1 (1.0). Each study had the following number of eccentricities: 1 (Clementz et al., 1994; Katsanis et al., 1997; Curtis et al., 2001a; Brownstein et al., 2003); 2 (McDowell and Clementz, 1997; Crawford et al., 1998); 3 (Thaker et al., 1996, 2000; Karoumi et al., 2001). d Number of trials: V 24 ( 1.0) vs. >25 (1.25). Each study had the following number of trials: 14 (Brownstein et al., 2003); 20 (Clementz et al., 1994; Katsanis et al., 1997; Curtis et al., 2001a); 24 (Crawford et al., 1998); 40 (McDowell and Clementz, 1997); 60 (Thaker et al., 1996; 2000; Karoumi et al., 2001). b

As Table 3 indicates, there is substantial variability in the effect sizes found in studies that used the overlap AS paradigm. The effect sizes for the Salt Lake City and Palau samples are much larger than the effect sizes for the San Diego sample (McDowell

et al., 1999). The effect sizes for the single peripheral target ( F 10j) used by Curtis et al. (2001b) are substantially smaller than the effect sizes for the ‘‘near’’ ( F 8j) peripheral target used by McDowell et al. (1999). We evaluated whether there was

118

D.L. Levy et al. / Schizophrenia Research 71 (2004) 113–125

Table 3 Antisaccade studies (overlap paradigm) of relatives of schizophrenia patients ordered chronologically Study

McDowell and Clementz (1997)a McDowell et al. (1999)a San Diego Salt Lake City Palau Curtis et al. (2001b)c

First-degree relatives of schizophrenia patients

Nonpsychiatric controls

Effect size

N

% Error score (near/far)

N

% Error score (near/far)

Cohen’s d (near/far)

r (near/far)

Glass’ delta (near/far)

60

10.0 (24.0)b

32

2.0 (11.0)b

0.39

0.19

0.73

60 29 41 42

31.0 48.0 47.0 12.3

94 94 94 38

22.0 22.0 22.0 9.9

0.43/0.70 1.23/1.97 1.30/1.90 0.23/NA

0.21/0.33 0.52/0.70 0.54/0.70 0.11/NA

0.50/1.375 1.44/3.75 1.39/3.5 0.29/NA

(25.0)/19.0 (23.0) (29.0)/38.0 (28.0) (22.0)/36.0 (24.0) (11.6)

(18.0)/8.0 (8.0) (18.0)/8.0 (8.0) (18.0)/8.0 (8.0) (8.4)

Mean (standard deviation). a Peripheral targets at F 8j (near), F 16j (far); San Diego RSP is the same group as in the 1997 study; error rate is not age-adjusted. b Age-adjusted error rates averaged across F 8j (near) and F 16j (far) targets. c Peripheral target at F 10j; error rate is not age-adjusted.

evidence of significant heterogeneity in these effect sizes using an extension of the Hedges and Olkin (1985) Q statistic. The Q statistic was developed to provide a test for heterogeneity when effect sizes are correlated because multiple variables or outcomes are measured on each subject. Cook (submitted for publication) has extended this method to estimate the correlation between effect sizes due to the nonindependence of the control group in the McDowell et al. (1999) study (three groups of RSP were compared with one control group), and used an estimate of this correlation in the Q statistic. The results of the heterogeneity analysis for the three samples in McDowell et al. (1999) were significant for both ‘‘near’’ (X2(2) = 22.95, P < 0.0001) and ‘‘far’’ targets (X2(2) = 134.53, P < 0.00000001), indicating that the RSP-control differences in the three comparisons for each target eccentricity were unlikely to be drawn from the same distribution. The analysis of the effect sizes for the ‘‘near’’ targets in four samples of RSP and controls (i.e., three from the McDowell et al., 1999 study and one from the Curtis et al., 2001b study) also showed significant heterogeneity (X2(3) = 27.25, P < 0.00005). We did not try to evaluate possible sources of effect size heterogeneity in studies that used the overlap paradigm, because of the small number of independent studies and control groups. Both studies used similarly asymmetric exclusion criteria, indicating that some other factor

must be the source of the heterogeneity in effect size.2 2.3. Gap AS task One study (Ross et al., 1998) used the gap version of the AS task to compare the performance of RSP 2

McDowell et al. (1999) have speculated that the much larger effect sizes in the Palau and Salt Lake City samples than in the San Diego sample may reflect differences in genetic loading for schizophrenia. They reasoned that all of the Palau and Salt Lake City families had multiple cases of schizophrenia, whereas all but one of the San Diego families had only one schizophrenic member. This explanation may account for variability in results based on the ‘‘overlap-far’’ AS paradigm, but it does not convincingly account for variability in results based on the standard or ‘‘overlap-near’’ AS paradigms. Using the standard AS paradigm to compare RSP from multiplex families with controls, Crawford et al. (1998) obtained a small effect size. Using the ‘‘overlap-near’’ AS paradigm to compare RSP primarily from simplex families with controls, Curtis et al. (2001b) obtained a much smaller effect size than McDowell et al. (1999) obtained for the San Diego families. Since submitting this manuscript, we have become aware of another study that compared the performance of RSP and controls on the standard AS task. The full study is currently unpublished, but the results have been presented as an abstract (MacCabe et al., 2002). The results showed that neither ‘‘obligate carrier’’ RSP, other relatives from multiplex families, nor relatives from families with only one schizophrenic member made significantly more errors than controls did. The effect sizes (Glass’ delta) ranged from 0.41 to 0.14. As in the Crawford et al. (1998) study, symmetrical exclusion criteria were applied to RSP and controls.

D.L. Levy et al. / Schizophrenia Research 71 (2004) 113–125

and controls. They divided eight ‘‘parental dyads’’ into two groups, one composed of the parent in each pair with a family history of schizophrenia (‘‘most likely gene carriers’’) and the other composed of the parent with no such family history (‘‘least likely gene carriers’’). The performance of each group was compared with that of controls. They found that the group of ‘‘most likely gene carrier’’ parents (but not the group of ‘‘least likely gene carrier’’ parents) made significantly more errors than controls did (Glass’ delta: 0.75; d: 0.81; r: 0.37).

3. Discussion Our results indicate that the effect sizes for comparisons of RSP and controls on the standard version of the AS task were significantly heterogeneous. The mean effect size is therefore not interpretable as a population indicator. Variability in effect size was accounted for by subject selection criteria but not by aspects of the experimental procedures. Significantly larger effect sizes were found in studies that applied less stringent exclusion criteria to RSP than to controls than in studies that applied comparable exclusion criteria to RSP and controls. Significant heterogeneity in effect size was also found for the overlap version of the AS task, but the specific sources of this heterogeneity could not be identified based on the existing literature. These findings, their implications for sample composition, and considerations relevant to a genetic model of AS performance are discussed below. 3.1. Asymmetric diagnostic exclusion criteria in controls and RSP Of the five studies that employed asymmetric diagnostic exclusion criteria (Clementz et al., 1994; McDowell and Clementz, 1997; Katsanis et al., 1997; Curtis et al., 2001a; Karoumi et al., 2001), four allowed psychiatric disorders in RSP but excluded those same disorders in the controls (Clementz et al., 1994; McDowell and Clementz, 1997; Katsanis et al., 1997; Curtis et al., 2001a). The fifth study (Karoumi et al., 2001) was asymmetric because although diagnostic exclusion criteria were symmetric with respect to personal psychopathology, they were asymmetric

119

with respect to familial psychopathology (an asymmetry that was also present in two studies that applied asymmetric personal psychopathology criteria). The effect sizes in these five studies were significantly larger than the effect sizes in the four studies that applied symmetrical exclusion criteria to both groups (Thaker et al., 1996, 2000; Crawford et al., 1998; Brownstein et al., 2003). Below we discuss each of these features of asymmetry. With respect to personal psychopathology, two kinds of asymmetry were present. The first involved excluding individuals with nonpsychotic Axis I disorders from the control group but not from the RSP group. In the Clementz et al. (1994) study, 7/32 RSP met criteria for past major depression, but controls with a major affective disorder were excluded. In addition, in the same study, three subjects who met diagnostic criteria for schizotypal personality disorder were included in the RSP group, but this condition was an exclusion criterion for controls. Similarly, in the Katsanis et al. (1997) and Curtis et al. (2001a) studies, 12/55 and 36/116 RSP, respectively, met diagnostic criteria for nonpsychotic mood and substance use disorders, which were exclusion criteria for controls. In the McDowell and Clementz (1997) study, controls with any Axis I disorder were excluded, but RSP with an Axis I disorder were not (see footnote b of Table 2). In the Karoumi et al. (2001) study, Axis I disorders were excluded from both RSP and controls, but controls were also excluded if there was a family history of any Axis I disorder, a topic that is discussed below. The same four studies that were asymmetric with respect to nonpsychotic Axis I disorders also included individuals with psychotic disorders in the RSP group (Clementz et al., 1994; McDowell and Clementz, 1997; Katsanis et al., 1997; Curtis et al., 2001a). Therefore, the magnitude of the contribution of each type of asymmetry to the effect size cannot be determined. There are strong a priori reasons, however, to expect that the presence of psychotic individuals in the RSP group will inflate the mean and variance of that group. The magnitude of these effects will, of course, depend on the proportion of RSP with psychotic disorders. The empirical literature clearly shows that the AS performance of individuals with psychotic conditions would be expected to increase the mean and variance of error

120

D.L. Levy et al. / Schizophrenia Research 71 (2004) 113–125

rate in the RSP group as a whole, because psychotic conditions, even in remission, are associated with an increased probability of poor performance. Every AS study of schizophrenia patients has reported significantly worse performance in these patients compared with controls (Fukushima et al., 1988, 1990a,b, 1994; Thaker et al., 1989; Rosse et al., 1993; Clementz et al., 1994; Matsue et al., 1994; Crawford et al., 1995a, 1998; Sereno and Holzman, 1995; Allen et al., 1996; Tien et al., 1996; Katsanis et al., 1997; McDowell and Clementz, 1997; Hutton et al., 1998; Karoumi et al., 1998, 2001; Levy et al., 1998; Maruff et al., 1998; Ross et al., 1998; McDowell et al., 1999; Muller et al., 1999; Brenner et al., 2001; Curtis et al., 2001a; Gooding and Tallent, 2001; Manoach et al., 2002). Moreover, schizophrenia patients in full remission perform as poorly as acutely psychotic schizophrenia patients (Curtis et al., 2001a). The mean error rate of the eight RSP with a history (but no current evidence) of psychosis was found to be ‘‘more similar’’ to the mean error rates of both acutely psychotic and remitted schizophrenia patients than to that of RSP with no Axis I pathology (Curtis et al., 2001a). In the Clementz et al. (1994) study, the relative with the highest error rate (more than 2 S.D. above the mean of RSP) was one of four RSP with a ‘‘schizophrenia-spectrum disorder’’. In the McDowell and Clementz (1997) study, the one relative with a diagnosis of schizophrenia had an error rate (on the overlap AS task) that was outside the range of the controls (performance of this individual on the standard AS task was not described). Similarly, both psychotic (Katsanis et al., 1997; Curtis et al., 2001a) and remitted bipolar patients (McDowell and Clementz, 1997; Gooding and Tallent, 2001) have been reported to have significantly elevated error rates compared with controls [in other studies bipolar patients did not perform more poorly than controls (Fukushima et al., 1990a; Clementz et al., 1994; Crawford et al., 1995a)]. With respect to family history of psychopathology, stricter exclusion criteria were applied to controls than to RSP in three of the studies with the largest effect sizes. In addition to the personal psychopathology screening criteria described above, controls were also excluded if family members had: (1) received treatment for a major affective disorder

or for substance abuse (Katsanis et al., 1997), (2) received any psychiatric treatment (Curtis et al., 2001a), or (3) any Axis I disorder (Karoumi et al., 2001). Thus, controls differed from RSP not only because they were not first-degree relatives of a schizophrenia patient, but also because they were not relatives of individuals with many other psychiatric disorders. As a result, the control groups remained more selective with respect to family history of psychiatric illness even when RSP without Axis I disorders were compared with controls (Katsanis et al., 1997; Curtis et al., 2001a; Karoumi et al., 2001). The four studies with the smallest effect sizes used the same personal and family history criteria to exclude both RSP and controls, with the exception that controls, unlike RSP, also had no family history of psychosis (Thaker et al., 1996, 2000; Crawford et al., 1998; Brownstein et al., 2003). In all four studies, psychotic individuals were excluded from both groups. In one study, neither group included individuals with Axis I disorders (Thaker et al., 2000) and in another both groups were largely free of Axis I disorders [Thaker et al. (1996) allowed in both groups a single episode of untreated major depression if it occurred more than 2 years earlier]. One study allowed nonpsychotic Axis I disorders in both groups and excluded individuals with schizophrenia-related personality disorders from both (Brownstein et al., 2003). One study allowed nonpsychotic disorders in both groups (Crawford et al., 1998). Our results indicate that applying more selective diagnostic exclusion criteria to RSP than to controls is a major condition for obtaining medium to large effect sizes in studies of RSP on the standard AS task. A conservative interpretation of this finding is that the larger effect sizes were not related to schizophrenia per se, but to the over-representation of psychiatric illnesses in RSP and the under-representation of the same disorders in the controls and in the relatives of the controls. This interpretation is consistent with reports that poor performance on the AS task is found not only in psychiatric conditions thought to be related to schizophrenia, such as psychometric and clinical schizotypy (Holzman et al., 1995; O’Driscoll et al., 1998; Gooding, 1999; Larrison et al., 2000; Brenner et al., 2001), but also

D.L. Levy et al. / Schizophrenia Research 71 (2004) 113–125

in a range of other psychiatric disorders. Findings vary for each of these diagnostic groups (for the sake of completeness we cite both positive and negative studies), but there is at least some support for increased error rate in patients with bipolar disorder (Sereno and Holzman, 1995; Crawford et al., 1995a,b; Katsanis et al., 1997; McDowell and Clementz, 1997; Curtis et al., 2001a; Gooding and Tallent, 2001), major depressive disorder (Katsanis et al., 1997; Sweeney et al., 1998; Curtis et al., 2001a), obsessive-compulsive disorder (Tien et al., 1992; McDowell and Clementz, 1997; Rosenberg et al., 1997; Maruff et al., 1999), and attention deficit hyperactivity disorder (Rothlind et al., 1991; Aman et al., 1998; Munoz et al., 1999, 2003). The presence of increased AS errors in a range of psychopathological conditions suggests that studies of AS performance may be less relevant for understanding schizophrenia per se than for understanding processes that are common to a broad range of psychiatric disorders. 3.2. Issues of sample composition It is clear that subject selection factors affect the magnitude of the performance difference between RSP and controls, an outcome that has more general methodological implications. The optimal inclusion/ exclusion criteria in any study, including the degree of symmetry and stringency of the exclusion criteria, depend on the goal of the study. For example, if one seeks to determine whether a particular process is associated with schizophrenia, it is inefficient to compare schizophrenia subjects with a control group that includes individuals with that disorder. Asymmetric exclusion criteria are thus appropriate in this case. If, however, one studies the same process in RSP in order to determine whether a particular behavior is useful as an auxiliary trait in linkage studies (i.e., a pleiotropic gene effect or an endophenotype), one must be able to distinguish a diathesis (the trait either causes the disease or is a pointer to an underlying causal process) from an epiphenomenon (a secondary effect caused by the disease) (Matthysse, 1993). Symmetrical exclusion criteria, in which both RSP and controls are purified of clinical conditions that could produce the trait as an epiphenomenon (Chapman and Chapman, 1973; Holzman

121

and Matthysse, 1990; Lenzenweger, 1998), are essential for making this distinction. In this case, symmetric exclusion criteria are appropriate. Purifying one group of potentially confounding conditions but not the other makes it likely that the groups will perform differently, but such a difference would not provide strong evidence that the trait is heritable, because it could be a secondary effect of the asymmetry in sample composition. 3.3. Differences in AS task administration The aspects of task administration that we examined did not contribute significantly to effect size heterogeneity. Specifically, neither number of target eccentricities nor number of trials was related to the magnitude of effect size. We had hypothesized that the number of trials used to assess AS performance might be related to the magnitude of RSP-control differences. For example, studies that used a larger number of trials might be expected to yield more stable estimates of performance ability than studies that used fewer trials. However, studies that administered relatively few trials (range: 14 – 24) (Clementz et al., 1994; Katsanis et al., 1997; Crawford et al., 1998; Curtis et al., 2001a; Brownstein et al., 2003) did not differ in effect size from those that administered a larger number of trials (range: 40 –60) (Thaker et al., 1996, 2000; McDowell and Clementz, 1997; Karoumi et al., 2001). Indeed, medium to large effect sizes were obtained in studies that used as few as 20 trials (Clementz et al., 1994; Katsanis et al., 1997; Curtis et al., 2001a), and small effect sizes were obtained in studies that used as many as 60 trials (Thaker et al., 1996, 2000). The nonsignificant effect of number of trials is consistent with the finding of stable performance across up to six blocks of 20 trials per block in schizophrenia, bipolar disorder, and obsessive-compulsive disorder patients as well as in nonpsychiatric controls (McDowell and Clementz, 1997). We had also hypothesized that task difficulty might increase with number of eccentricities, which might be associated with larger effect sizes. However, number of eccentricities was not a significant source of variability in effect size. Medium to large effect sizes were obtained in studies that used only one eccentricity (Clementz et al., 1994; Katsanis et al., 1997; Curtis et al., 2001a) and small effect sizes were obtained in

122

D.L. Levy et al. / Schizophrenia Research 71 (2004) 113–125

studies that used multiple eccentricities (Thaker et al., 1996, 2000). We do not rule out the possibility that variables relevant to test administration do play a role in effect size heterogeneity. For example, we were unable to assess the influence of variables that were not consistently reported but which could have had an effect, such as inter-trial interval, whether subjects were given feedback on performance during the experimental trials, and number of practice trials. 3.4. Confidence limits It is interesting to note that if the underlying sources of heterogeneity had been unknown, an analysis based on the confidence limits alone would have revealed that a systematic difference between RSP and controls, if there is one, must be small, but a large effect could be ruled out. That result would not represent the underlying reality accurately, because the heterogeneity analysis shows that there is not a consistent small effect.3 Rather, we have a collection of studies, some of which show large effects and some of which show small effects, with the magnitude of

3

Using a normal theory approximation, the upper 95% confidence limit for the difference between the mean error scores of RSP and relatives of nonpsychiatric controls in the Brownstein et al. (2003) study is 6.31, indicating that a mean difference between the groups smaller than 6.31 cannot be ruled out, even though the observed mean difference was 1.0. The upper 95% confidence limits for the three other studies whose effect sizes were small (Thaker et al., 1996, 2000; Crawford et al., 1998) are 7.64, 10.50, and 15.22, respectively, showing substantial overlap with the differences in means reported in the studies with medium – large effect sizes. These findings indicate that the studies reporting small effect sizes do not rule out a small effect, or an effect in a small proportion of RSP, but they do rule out a consistent large group effect. When applying this procedure in the converse way, the lower confidence limits of the studies with medium – large effect sizes were 1.82, 3.48, 5.97, and 9.25, respectively (Clementz et al., 1994; McDowell and Clementz, 1997; Karoumi et al., 2001; Curtis et al., 2001a), barely overlapping the upper confidence intervals of the studies with small effect sizes [the Katsanis et al. (1997) study was excluded, because it presented medians and interquartile ranges, rather than means and standard deviations]. Therefore, using this method, the confidence intervals for the two groups of studies have sufficient overlap that it is not possible to rule out a consistent small effect underlying all of the findings, but a consistent large effect can be ruled out. The meta-analysis, however, does rule out a consistent small effect.

the effects varying as a function of the subject selection criteria used. 3.5. Expectations from a genetic model As indicated earlier, the finding of increased errors on the AS task in schizophrenia patients and RSP has led to the speculation that AS performance may tap processes related to genetic vulnerability. According to a genetic model that involves co-familial transmission, RSP would be expected to have both a higher mean error rate and a larger variance than controls. As we showed above, however, the magnitude of the observed mean difference between RSP and controls varies as a function of subject selection criteria. From the data presented in published studies, it is possible to test whether RSP had a significantly larger variance than controls. The variance ratios for the eight independent studies that provided means and standard deviations of RSP and controls are shown in Table 4. Under the null hypothesis the variance ratio is 1 and is distributed as F. A test of whether the variances of RSP were larger than those of controls was not statistically significant ( F values were converted to Table 4 Variance ratiosa in antisaccade studies (standard paradigm) of firstdegree relatives of schizophrenia patients (ordered from smallest to largest) Study

Variance ratiob

P-value

ln(variance ratio)c

Thaker et al. (1996) Brownstein et al. (2003) Karoumi et al. (2001) Thaker et al. (2000) Crawford et al. (1998) McDowell and Clementz (1997) Curtis et al. (2001a) Clementz et al. (1994)

F = 0.32, df = 25,67 F = 0.90, df = 97,23 F = 1.40, df = 20,20 F = 1.54, df = 54,61 F = 1.59, df = 49,37 F = 1.72, df = 59,31

>0.995

1.14

>0.50

0.10

0.10 < P < 0.25

0.33

0.05 < P < 0.10

0.43

0.05 < P < 0.10

0.46

0.05 < P < 0.10

0.54

0.01 < P < 0.025

0.54

0.025 < P < 0.05

0.76

a 2

F = 1.72, df = 115,108 F = 2.13, df = 31,32

s RSP/s2NORMALS (one-sided to the right). Mean: 1.4; standard deviation: 0.56; median: 1.6. c Median: 0.445. b

D.L. Levy et al. / Schizophrenia Research 71 (2004) 113–125

the natural log of F so that studies with F’s above and below 1.0 would be equally weighted: Wilcoxon Signed Rank Test; S = 9.0, P = 0.11, one sided). From the meta-analysis, we conclude that the use of a purified control group yields large and statistically significant RSP-control effects, whereas applying exclusion criteria of comparable stringency in the two groups yields small and statistically nonsignificant effects. On the basis of the studies in the literature, therefore, we conclude that performance on the standard AS task does not appear to be a pleiotropic effect of a schizophrenia gene. For the purpose of detecting the presence of a pleiotropic effect of a schizophrenia gene, applying criteria of comparable stringency in the RSP and control groups is the proper strategy. Using asymmetrical inclusion/exclusion criteria is bound to create an upward bias in the direction of magnifying group differences (Smith and Iacono, 1986; Tsuang et al., 1988; Schwartz and Link, 1989; Kendler, 1990), and may tend falsely to suggest an effect related to schizophrenia gene. In an ideal genetic study, two groups are compared that differ only in the probability that members of the groups have a gene relevant to schizophrenia and not in the presence of unrelated psychopathology. Although perfect symmetry of inclusion/exclusion criteria typically cannot be achieved (i.e., in the absence of definitive diagnostic information, a family history of psychosis will usually exclude a control but not an RSP), reasonable symmetry can be achieved. The significant within-family correlation in performance, both in RSP (Crawford et al., 1998; Curtis et al., 2001a; Brownstein et al., 2003) and in twins who were not ascertained for being RSP (Malone and Iacono, 2002), suggests that genetic effects may contribute to variability in AS performance, but in order to show convincingly that this trait is related to a schizophrenia genotype, RSP must be shown to perform worse than controls in samples ascertained on the basis of comparable diagnostic exclusion criteria.

Acknowledgements This study was supported in part by NIMH grants MH49487, MH31340, MH01021, MH31154, by a grant from The Roy Hunt Foundation, by an operating grant from the Canadian Institute of Health Research,

123

and by a Harvard University Graduate School of Arts and Sciences Merit Fellowship. The authors thank Donald Rubin for his helpful comments. We also thank Drs. Thierry d’Amato, Monica Calkins, Brett Clementz, Jennifer McDowell, and Gunvant Thaker for clarifying methodological questions. The method developed by Cook to handle correlated effect sizes is available on request. References Allen, J.S., Lambert, A.J., Attah Johnson, F.Y., Schmidt, K., Nero, K.L., 1996. Antisaccadic eye movements and attentional asymmetry in schizophrenia in three Pacific populations. Acta Psychiatr. Scand. 94, 258 – 265. Aman, C.J., Roberts, R.J., Pennington, B.F., 1998. A neuropsychological examination of the underlying deficit in attention deficit hyperactivity disorder: frontal lobe versus right parietal lobe theories. Dev. Psychol. 34, 956 – 969. Brenner, C.A., McDowell, J.E., Cadenhead, K.S., Clementz, B.A., 2001. Saccadic inhibition among schizotypal personality disorder subjects. Psychophysiology 38, 399 – 403. Brownstein, J., Krastoshevsky, O., McCollum, C., Kundamal, S., Matthysse, S., Holzman, P.S., Mendell, N.R., Levy, D.L., 2003. Antisaccade performance is abnormal in schizophrenia patients but not in their biological relatives. Schizophr. Res. 63, 13 – 15. Chapman, L.J., Chapman, J.P., 1973. Problems in the measurement of cognitive deficit. Psychol. Bull. 79, 380 – 385. Clementz, B.A., McDowell, J.E., Zisook, S., 1994. Saccadic system functioning among schizophrenia patients and their first-degree relatives. J. Abnorm. Psychology 103, 277 – 287. Cohen, J., 1977. Statistical Power Analysis for the Behavioral Sciences. Academic Press, New York. Cook, S., 2004. A note on testing for heterogeneity among correlated effect sizes (submitted for publication). Crawford, T.J., Haegar, B., Kennard, C., Reveley, M.A., Henderson, L., 1995a. Saccadic abnormalities in psychotic patients: I. Neuroleptic-free psychotic patients. Psychol. Med. 25, 461 – 471. Crawford, T.J., Haegar, B., Kennard, C., Reveley, M.A., Henderson, L., 1995b. Saccadic abnormalities in psychotic patients: II. The role of neuroleptic treatment. Psychol. Med. 25, 473 – 483. Crawford, T.J., Sharma, T., Puri, B.K., Murray, R.M., Lewis, S.W., 1998. Saccadic eye movements in families multiply affected with schizophrenia: the Maudsley family study. Am. J. Psychiatry 155, 1703 – 1710. Curtis, C.E., Calkins, M.E., Grove, W.M., Feil, K.J., Iacono, W.G., 2001a. Saccadic disinhibition in patients with acute and remitted schizophrenia and their first-degree biological relatives. Am. J. Psychiatry 158, 100 – 106. Curtis, C.E., Calkins, M.E., Iacono, W.G., 2001b. Saccadic disinhibition in patients and their first-degree biological relatives. Exp. Brain Res. 137, 228 – 236. Fischer, B., Weber, H., 1992. Characteristics of ‘‘anti’’ saccades in man. Exp. Brain Res. 89, 415 – 424. Fischer, B., Weber, H., 1997. Effects of stimulus conditions on

124

D.L. Levy et al. / Schizophrenia Research 71 (2004) 113–125

the performance of antisaccades in man. Exp. Brain Res. 116, 191 – 200. Fukushima, J., Fukushima, K., Chiba, T., Tanaka, S., Yamashita, I., Kato, M., 1988. Disturbances of voluntary control of saccadic eye movements in schizophrenic patients. Biol. Psychiatry 23, 670 – 677. Fukushima, J., Morita, N., Fukushima, K., Chiba, T., Tanaka, S., Yamashita, I., 1990a. Voluntary control of saccadic eye movements in patients with schizophrenic and affective disorders. J. Psychiatr. Res. 24, 9 – 24. Fukushima, J., Fukushima, K., Morita, N., Yamashita, I., 1990b. Further analysis of the control of voluntary saccadic eye movements in schizophrenic patients. Biol. Psychiatry 28, 943 – 958. Fukushima, J., Fukushima, K., Miyasaka, K., Yamashita, I., 1994. Voluntary control of saccadic eye movement in patients with frontal cortical lesions and parkinsonian patients in comparison with that in schizophrenics. Biol. Psychiatry 36, 21 – 30. Glass, G.V., 1976. Primary, secondary, and meta-analysis of research. Educ. Res. 5, 3 – 8. Gooding, D.C., 1999. Antisaccade task performance in questionnaire-identified schizotypes. Schizophr. Res. 35, 157 – 166. Gooding, D.C., Tallent, K.A., 2001. The association between antisaccade task and working memory task performance in schizophrenia and bipolar disorder. J. of Nerv. Ment. Dis. 189, 8 – 16. Greenberg, D.A., Durner, M., Keddache, M., Shinnar, S., Resor, S.R., Moshe, S.L., Rosenbaum, D., Cohen, J., Harden, C., Kang, H., Wallace, S., Luciano, D., Ballaban-Gil, K., Tomasini, L., Zhou, G., Klotz, I., Dicker, E., 2000. Reproducibility and complications in gene searches: linkage on chromosome 6, heterogeneity, association, and maternal inheritance in juvenile myoclonic epilepsy. Am. J. Hum. Genet. 66, 508 – 516. Hedges, L.V., Olkin, I., 1985. Statistical Methods for Meta-Analysis. Academic Press, Orlando, FL. Holzman, P.S., Matthysse, S., 1990. The genetics of schizophrenia: a review. Psychol. Sci. 1, 279 – 286. Holzman, P.S., Coleman, M., Lenzenweger, M.F., Levy, D.L., Matthysse, S., O’Driscoll, G., Park, S., 1995. Working memory deficits, antisaccades, and thought disorder in relation to perceptual aberration. In: Raine, A., Lencz, T., Mednick, S. (Eds.), Schizotypal Personality. Cambridge Univ. Press, New York, pp. 353 – 381. Hutton, S.B., Crawford, T.J., Puri, B.K., Duncan, L.J., Chapman, M., Kennard, C., Barnes, T.R.E., Joyce, E.M., 1998. Smooth pursuit and saccadic abnormalities in first-episode schizophrenia. Psychol. Med. 28, 685 – 692. Karoumi, B., Ventre-Dominey, J., Vighetto, A., Dalery, J., d’Amato, T., 1998. Saccadic eye movements in schizophrenic patients. Psychiatry Res. 77, 9 – 19. Karoumi, R., Saoud, M., d’Amato, T., Rosenfeld, F., Denise, P., Gutknecht, C., Gaveau, V., Beaulieu, F., Dalery, J., Rochet, T., 2001. Poor performance in smooth pursuit and antisaccadic eyemovement tasks in healthy siblings of patients with schizophrenia. Psychiatry Res. 101, 209 – 219. Katsanis, J., Kortenkamp, S., Iacono, W.G., Grove, W.M., 1997. Antisaccade performance in patients with schizophrenia and affective disorder. J. Abnorm. Psychology 106, 468 – 472. Kendler, K.S., 1990. The super-normal control group in psychiatric

genetics: possible artifactual evidence for coaggregation. Psychiatr. Genet. 1, 45 – 53. Kendler, K.S., McGuire, M., Gruenberg, A., O’Hare, A., Spellman, M., Walsh, D., 1993. The Roscommon family study: I. Methods, diagnosis of probands, and risk of schizophrenia in relatives. Arch. Gen. Psychiatry 50, 527 – 540. Larrison, A.L., Ferrante, C.F., Briand, K.A., Sereno, A.B., 2000. Schizotypal traits, attention and eye movements. Prog. Neuropsychopharmacol. Biol. Psychiatry 24, 357 – 372. Lenzenweger, M.F., 1998. Schizotypy and schizotypic psychopathology: mapping an alternative expression of schizophrenia liability. In: Lenzenweger, M.F., Dworkin, R. (Eds.), Origins and Development of Schizophrenia: Advances in Experimental Psychopathology. American Psychological Association Press, Washington, DC, pp. 93 – 121. Levy, D.L., Mendell, N.R., LaVancher, C., Brownstein, J., Shorrock, K., Krastoshevsky, O., Teraspulsky, L., Lo, Y., Bloom, R., Matthysse, S., Holzman, P.S., 1998. Disinhibition in antisaccade performance in schizophrenia. In: Lenzenweger, M.F., Dworkin, R. (Eds.), Origins and Development of Schizophrenia: Advances in Experimental Psychopathology. American Psychological Association Press, Washington, DC, pp. 185 – 210. MacCabe, J.H., Simon, H., Murray, J.M., 2002. Is antisaccade performance a marker of genetic liability in schizophrenia. Schizophr. Res. 53, 78 – 79. Malone, S.M., Iacono, W.G., 2002. Error rate on the antisaccade task: heritability and developmental change in performance among preadolescent and late-adolescent female twin youth. Psychophysiology 39, 664 – 673. Manoach, D.S., Lindgren, K.A., Cherkasova, M.V., Goff, D.C., Halpern, E.F., Intriligator, J., Barton, J.J.S., 2002. Schizophrenic subjects show deficient inhibition but intact task switching on saccadic tasks. Biol. Psychiatry 51, 816 – 826. Maruff, P., Danckert, J., Pantelis, C., Currie, J., 1998. Saccadic and attentional abnormalities in patients with schizophrenia. Psychol. Med. 28, 1091 – 1100. Maruff, P., Purcell, R., Tyler, P., Pantelis, C., Currie, J., 1999. Abnormalities of internally generated saccades in obsessivecompulsive disorder. Psychol. Med. 29, 1377 – 1385. Matsue, Y., Saito, H., Osakabe, K., Awata, S., Ueno, T., Matsuoka, H., Chiba, H., Fuse, Y., Sato, M., 1994. Smooth pursuit eye movements and voluntary control of saccades in the antisaccades task in schizophrenic patients. Jap. J. Psychiatry Neurol. 48, 13 – 22. Matthysse, S., 1993. Genetics and the problem of causality in abnormal psychology. In: Sutker, P.B., Adams, H.E. (Eds.), Comprehensive Handbook of Psychopathology. Plenum, New York, NY, pp. 47 – 56. Matthysse, S., Parnas, J., 1992. Extending the phenotype of schizophrenia: implications for linkage analysis. J. Psychiatr. Res. 26, 329 – 344. McDowell, J.E., Clementz, B.A., 1997. The effect of fixation condition manipulations on antisaccade performance in schizophrenia: studies of diagnostic specificity. Exp. Brain Res. 115, 333 – 344. McDowell, J.E., Myles-Worsley, M., Coon, H., Byerley, W., Clementz, B.A., 1999. Measuring liability for schizophrenia using

D.L. Levy et al. / Schizophrenia Research 71 (2004) 113–125 optimized antisaccade stimulus parameters. Psychophysiology 36, 138 – 141. Muller, N., Riedel, M., Eggert, T., Straube, A., 1999. Internally and externally guided voluntary saccades in unmedicated and medicated schizophrenics: II. Saccadic latency, gain, and fixation suppression errors. Eur. Arch. Psychiatry Clin. Neurosci. 249, 7 – 14. Munoz, D.P., Hampton, K.A., Moore, K.D., Goldring, J.E., 1999. Control of purposive saccadic eye movements and visual fixation in children with attention-deficit hyperactivity disorder. In: Becker, W., Deubel, H., Mergner, T. (Eds.), Current Oculomotor Research: Physiological and Psychological Aspects. Plenum, New York, pp. 415 – 423. Munoz, D.P., Armstrong, I.T., Hampton, K.A., Moore, K.D., 2003. Altered control of visual fixation and saccadic eye movements in attention-deficit hyperactivity disorder. J. Neurophysiol. 90, 503 – 514. O’Driscoll, G.A., Lenzenweger, M.F., Holzman, P.S., 1998. Antisaccades and smooth pursuit eye movements and schizotypy. Arch. Gen. Psychiatry 55, 837 – 843. Rosenberg, D.R., Averbach, D.H., O’Hearn, K.M., Seymour, A.B., Birmaher, B., Sweeney, J.A., 1997. Oculomotor response inhibition abnormalities in pediatric obsessive-compulsive disorder. Arch. Gen. Psychiatry 54, 831 – 838. Rosenthal, R., 1984. Meta-analytic Procedures for Social Research. Sage Publications, Beverly Hills, CA. Rosenthal, R., Rosnow, R.L., 1984. Essentials of Behavioral Research: Methods and Data Analysis. McGraw-Hill, New York, NY. Rosenthal, R., Rubin, D.B., 1982. Comparing effect sizes of independent studies. Psychol. Bull. 92, 500 – 504. Ross, R.G., Harris, J.G., Olincy, A., Radant, A., Adler, L.E., Freedman, R., 1998. Familial transmission of two independent saccadic abnormalities in schizophrenia. Schizophr. Res. 30, 59 – 70. Rosse, R.B., Schwartz, B.L., Kim, S.Y., Deutsch, S.I., 1993. Correlation between antisaccade and Wisconsin card sorting test performance in schizophrenia. Am. J. Psychiatry 150, 333 – 335.

125

Rothlind, J.C., Posner, M.I., Schaughency, E.A., 1991. Lateralized control of eye movements in attention deficit hyperactivity disorder. J. Cogn. Neurosci. 3, 377 – 381. Schwartz, S., Link, B.G., 1989. The ‘well control’ artefact in case/ control studies of specific psychiatric disorders. Psychol. Med. 19, 737 – 742. Sereno, A.B., Holzman, P.S., 1995. Antisaccades and smooth pursuit eye movements in schizophrenia. Biol. Psychiatry 37, 394 – 401. Smith, G.N., Iacono, W.G., 1986. Lateral ventricular size in schizophrenia and choice of control group. Lancet 1, 1450. Sweeney, J.A., Strojwas, M.H., Mann, J.J., Thase, M.E., 1998. Prefrontal and cerebellar abnormalities in major depression: evidence from oculomotor studies. Biol. Psychiatry 43, 584 – 594. Thaker, G.K., Nguyen, J.A., Tamminga, C.A., 1989. Increased saccadic distractibility in tardive dyskinesia: functional evidence for subcortical GABA dysfunction. Biol. Psychiatry 25, 49 – 59. Thaker, G.K., Cassady, S.L., Adami, H.M., Moran, M., Ross, D.E., 1996. Eye movements in spectrum personality disorders: comparison of community subjects and relatives of schizophrenic patients. Am. J. Psychiatry 153, 362 – 368. Thaker, G.K., Ross, D.E., Cassady, S.L., Adami, H.M., Medoff, D.R., Sherr, J., 2000. Saccadic eye movement abnormalities in relatives of patients with schizophrenia. Schizophr. Res. 45, 235 – 244. Tien, A.Y., Pearlson, G.D., Machlin, S.R., Bylsma, F.W., HoehnSarie, R., 1992. Oculomotor performance in obsessive-compulsive disorder. Am. J. Psychiatry 149, 641 – 646. Tien, A.Y., Ross, D.E., Pearlson, G., Strauss, M.E., 1996. Eye movements and psychopathology in schizophrenia and bipolar disorder. J. of Nerv. Ment. Dis. 184, 331 – 338. Tsuang, M.T., Fleming, J.A., Kendler, K.S., Gruenberg, A.S., 1988. Selection of controls for family studies. Biases and implications. Arch. Gen. Psychiatry 45, 1006 – 1008.